Security context

What an agent needs to avoid regressing past fixes and find the next vuln in this repo.

evalstate/fast-agent
main @ 2f43843
4
Fixes
0
CVEs
HIGH
Peak severity
0.0%
Coverage
Highlights
Auth Bypass: 3 prior fixes. Scrutinize any change in this area.
mcp_registry.py: most-fixed (1 issue). Treat as high-risk during review.
3 high-severity fixes in this history; regressions here are high-impact.
Recurring patterns

The bug types that recur here, drawn from past fixes, not open vulnerabilities.

Auth Bypass: Remote-origin MCP skills can self-escalate privileges by containing permission-widening frontmatter keys such as 'allowed-tools' and 'hooks' during registry installation. Failure to strip these parameters allows untrusted skills to execute arbitrary privileged actions.
Auth Bypass: Untrusted agents can perform unauthorized local file reads, writes, and shell execution if permission checks and user consent gates are missing or bypassed within ACP filesystem runtimes.
Path Traversal: Insecure directory extraction of skill archives can allow directory traversal, particularly using Windows backslashes and drive anchors, resulting in arbitrary file write outside the intended skill directory.